Skip to content

Latest commit

 

History

History
56 lines (45 loc) · 3.64 KB

README.md

File metadata and controls

56 lines (45 loc) · 3.64 KB
Raw

HPC Conda Environments (hpc-conda-envs)

This repository provides conda environment files and matching LMOD files that are generally useful in a shared multi-user HPC cluster environment. The repository provide detailed instruction when support Jupyter Notebooks under Open OnDemand(OOD).

User Created Environments

This repository provides conda environments intended to be installed and admininstred by system administrators but users are able to leverage this installation to build their own custom environments. The custom environments will be created in the users home directory but symlinks to used to point to the system package cache reducing disk space usage and network transfer requirements.

To use a custom environment as a Jupyter kernel, only the appropriate kernel package (i.e. ipykernel, r-irkernel) needs to be installed in the environment. Do not include all the Jupyter packages. These are provided by system managed Jupyter environment.

Integration with LMOD

Users are able to use either module load or conda activate to activate the conda environments. The following two commands would be considered equivalent.

module load conda/envs/jupyter 
conda activate jupyter

Installation

The installation assumes that LMOD has already been installed on the system.

  1. Install miniconda3 at a shared location such as /usr/local/miniconda3. Answer "no" when asked to initialize Miniconda3.
  2. Activate conda's base environment using the 'eval' command shown following the install.
  3. (optional -- recommended) Add mamba to the base environment using conda install -c conda-forge mamba. Mamba is a fast drop-in alternative to conda.
  4. Edit the modules/miniconda3.lua file to point to the local miniconda3 installation.
  5. Examine the files in module/conda/envs. These will be the system installed conda environments. Create any other custom conda environments required on the local system. Optionally, add module files to modules/conda/envs provide conda environment activation using LMOD.
  6. Recursively build the conda environments within this repository by running find modules -name \*.yml -exec mamba env create -f {} \; from the root of this repo. This will build multiple conda environments and store them under the local miniconda3 installation.
  7. Recursively copy every thing under the repository modules directory into the system modules directory using cp -r modules/* <system_module_directory>.
  8. (optional) Create any other custom conda environments required on the local system. You can add module files to modules/conda/envs if you wish to provide environment activation using modules.
  9. Recursively fix permissions on the ENTIRE miniconda3 installation and conda modules with chmod -R g-w,o+rX /usr/local/miniconda3 /usr/local/modulefiles/conda.

Remember to fix permission on the miniconda3 installation after any changes to conda environments.

The Jupyter Environment

One common use case for this setup is to support Jupyter Notebooks under Open OnDemand (OOD). In this case, a single environment provides the necessary Jupyter package and plugins to support the web interface. Other conda environments provide the kernel used to execute the notebook. The kernel environments can be selected interactively from Jupyter interface.

This approach provides several advantages. First, the Jupyter environment and plugins are managed by system administrators and provide a consistent interface to all users. Second, commonly used kernel environments can be provided to users in a consistent and documented manner. Third, users are still able to create their own kernels for custom computing needs. (See the section on User Environments).